Real-Time Face Detection Using Multiple Instances Boosting Cascade

Considering the conventional defects of boosting cascade, such as overwhelmed training computation, inaccurate threshold adjustment, face detection based on multiple instances and boosting cascade presents a new way. This paper explores the solutions of boosting machine learning and its threshold adjustment strategies, which utilize separated training sets, large scale set with bootstrapping, various parameters adjustment, and multiple instance threshold adjustment. After applying these strategies, our method greatly improve the speed and accuracy of face detection, and explicitly upgrade the inside methodology of conventional boosting cascade face detection.